[offtopic]
I think I asked once, but didn't get any answer (or if I missed, please don't blame me, and just give a short yes/no answer): Any benefit for non-haswell guys to upgrade to v28?
[/offtopic]

[QUOTE=LaurV;352789][offtopic]
I think I asked once, but didn't get any answer (or if I missed, please don't blame me, and just give a short yes/no answer): Any benefit for non-haswell guys to upgrade to v28?
[/offtopic][/QUOTE]


No.

Thanks a lot.
(it works for me, no time for upgrades right now, really busy here around).

[QUOTE=TheMawn;352689]Are you using two or four 140mm fans? Either way you should be getting a LOT more cooling power than me. Lower frequency and same voltage, and still getting ~10C above me?[/QUOTE]

I'm using 3x140mm fans. The fourth one doesn't fit as the motherboard power cable is in the way. The cooling capacity is top end, but the design of the Haswell (4xxx) still makes it run way hotter than an Ivybridge (3xxx).

On Haswell, the voltage regulators are now integrated onto the CPU. As a result, most people are reporting ~10 degrees higher than an Ivybridge at the same speed. Exactly inline comparing your system to mine.

I'm told this allows Intel to more precisely control voltages and sleep states. For laptops, tablets and PCs that spend a lot of time idle, this reduces power consumption noticably, allowing their mobile chips to get within striking range of ARM based processors. But for our purposes where the CPU is at 100% utilisation all the time, it's a step backwards - limits overclocking and requires exotic cooling.

[QUOTE=TheMawn;352689]What memory do you use?[/QUOTE]

Currently using Corsair XMS3 2000MHz, 4GBx2.
Seeing that P95 is bandwidth limited, especially now with FMA3, thinking of upgrading this to a 2400MHz 8GBx2.

[QUOTE=henryzz;352726]Do you get any performance boost from running the core and clock at the same speed? I would imagine that if the cpu recognized that and locked them together then it could provide some improvement. Similar things like running memory at 2x fsb have been reckoned to provide small boosts in the past.[/QUOTE]

Not that I have noticed... when running 4 threads, the bandwidth limitation from the DDR3 2000MHz RAM ends up giving me the exact same iteration times whether the cache is 200MHz slower or not.

The general consensus from the overclocking forums is locking the 2 speeds has only a small benefit. It's better to set the cache speed around 200-300MHz lower than the core clockspeed. This will allow you to focus on overclocking the core more, and the gains there will be far more noticable than locking the 2 to the same speed.

[QUOTE=TheMawn;352745]I just caught this. Wasn't really paying attention. Just saw 1.275 V 4.2 GHz 1.26 V 4.0 GHz.
I have never heard of a cache speed...[/QUOTE]

For a while now (since Nehalem if I recall correctly), the Intel CPUs have been split into several domains, which can be run at different speeds. These are the "core" and the "uncore". The "uncore" contains things like the L3 cache, memory controller etc.

Recently it gots even more complicated with integratred graphics added and again running at a different speed. The domains are all connected to a ring bus, so each are quite separate and can have independant speeds and voltages (and hence power budgets) from one another.

The ring bus is a very elegant solution compared to having a fixed set of multipliers to choose from (e.g. the 2x FSB mentioned earlier). Especially since there are multiple domains now, each able to dynamically clock down to save power when idle.

This is just my understanding as an engineer (but I'm from mechanical, so someone please correct me if mistaken).

[QUOTE=db597;352798]This is just my understanding as an engineer (but I'm from mechanical, so someone please correct me if mistaken).[/QUOTE]
No need, you described very accurate what is going on. :tu:

Have we anyone here with a GT3e igpu? Benchmarks of thread scaling would be interesting. They have been out a bit now.

If you can't run v28 on your haswell+OC setup stably, just revert to v27 - the speed difference is not worth losing major runtime due to system instability.

I expect some people will find they can OC their haswell more aggressively using the low-MUL v27, which would make up much if not all the per-cycle performance difference which comes from using FMA.

[QUOTE=ewmayer;352843]If you can't run v28 on your haswell+OC setup stably, just revert to v27 - the speed difference is not worth losing major runtime due to system instability.

I expect some people will find they can OC their haswell more aggressively using the low-MUL v27, which would make up much if not all the per-cycle performance difference which comes from using FMA.[/QUOTE]

I don't agree, at least on my Haswell, v28.1 faster than v27.9:

[QUOTE=TheJudger;351553]
4000MHz 1200mV vCore, DDR3-2133 CL11
Copied worktodo from one of my machines doing 4 doublechecks, 3 times "FMA3 FFT length 1600K, Pass1=640, Pass2=2560", 1 time "FMA3 FFT length 1728K, Pass1=384, Pass2=4608":

v27.9: 10.7xx ms/iter for 1600k FFT, 10.5xx ms/iter for 1728k FFT, package power ~88W
v28.1: 9.1xx ms/iter for 1600k FFT, 10.0xx ms/iter for 1728k FFT, package power ~97W
Linpack: up to 123-128W package power

Oliver[/QUOTE]

v27.9 had choosen other Pass1 and Pass2 settings for 1600k FFT length.

Please ignore voltage and power consumption, I'm currently running 4000MHz at 1250mV, 1200mV isn't stable for Linpack.

Oliver

[QUOTE=TheJudger;352851]I don't agree, at least on my Haswell, v28.1 faster than v27.9[/QUOTE]

Do your timings imply that you are able to run v28 stably?